Download Assembling a nutrient database for a large cohort study: Blue

University of Wollongong
Research Online
Faculty of Health and Behavioural Sciences - Papers
(Archive)
Faculty of Science, Medicine and Health
2008
Assembling a nutrient database for a large cohort
study: Blue Mountains Eye Study
Victoria M. Flood
University of Wollongong, [email protected]
W Smith
Centre for Clinical Epidemiology, Uni of Newcastle
E Rochtchina
Centre for Vision Research, Uni of Sydney
Jie J. Wang
Centre for Vision Research, Uni of Sydney
Paul Mitchell
Centre for Vision Research, Uni of Sydney
Publication Details
Flood, V. M., Smith, W., Rochtchina, E., Wang, J. J. & Mitchell, P. 2008, 'Assembling a nutrient database for a large cohort study: Blue
Mountains Eye Study', Food Australia, vol. 60, no. 1-2, pp. 37-40.
Research Online is the open access institutional repository for the University of Wollongong. For further information contact the UOW Library:
[email protected]
Assembling a nutrient database for a large cohort study: Blue Mountains
Eye Study
Abstract
To construct a nutrient database for use with a large population-based cohort study, the Blue Mountains Eye
Study, dietary intakes were estimated using the Australian electronic food composition nutrient databases to
which values for additional nutrients were added, including folate, vitamin B12, carotenoids and fatty acids.
The addition of nutrient data from alternative, overseas, database sources may be useful in relation to the
assessment of outcome measures, however the data obtained from these sources need to be interpreted with
care, especially in relation to absolute quantities of intake.
Keywords
eye, assembling, nutrient, database, study, blue, cohort, large, mountains
Disciplines
Arts and Humanities | Life Sciences | Medicine and Health Sciences | Social and Behavioral Sciences
Publication Details
Flood, V. M., Smith, W., Rochtchina, E., Wang, J. J. & Mitchell, P. 2008, 'Assembling a nutrient database for a
large cohort study: Blue Mountains Eye Study', Food Australia, vol. 60, no. 1-2, pp. 37-40.
This journal article is available at Research Online: http://ro.uow.edu.au/hbspapers/344
Assembling a nutrient database
for a large cohort study: the
Blue Mountains Eye Study
V. M. Flood, W. Smith, E. Rochtchina,
J. J. Wang and P. Mitchell
To construct a nutrient database for use with
a large population-based cohort study, the Blue
Mountains Eye Study, dietary intakes were estimated
using the Australian food composition electronic
nutrient databases to which values for additional
nutrients were added, including folate, vitamin
B12, carotenoids, and fatty acids. The addition of
nutrient data from alternative, overseas, database
sources may be useful in relation to the assessment
of outcome measures, however the data obtained
from these sources needs to be interpreted with
care, especially in relation to absolute quantities of
intake.
Development of a nutrient database for a population
cohort study is fraught with difficulties, given that the
nutrient data is collected over a long period of time and
both the food supply and laboratory analysis methods
may change during that period. In addition, research
questions of interest may change over time, and may
therefore require additional information of particular
nutrients that were not available or not anticipated in
the original baseline data collection period. There are
also practical considerations in relation to storage of
data, construction of the nutrient database, and software
used for the analyses of the dietary assessment tool. The
database needs to be flexible enough to meet the current
and the future needs of researchers involved with a
longitudinal study.
This paper reflects personal experiences and practical
considerations in relation to these issues, based on
involvement with a population-based cohort study over
ten years.
The study population
The Blue Mountains Eye Study (BMES) is a populationbased cohort study of eye diseases and other chronic
health outcomes in residents aged 49 years and over of
a defined area, west of Sydney, Australia. Participants
were first examined during the period January 1992 to
January 1994. A door knock census of 38 ABS census
collection districts (CCD) in two adjoining postcode
areas (2780 and 2782) was conducted. Of 4433 eligible
residents, 3654 aged 49 to 97 years, attended an eye
examination (response rate 82.4%). Prior to attending
the eye examination, participants were sent a 145 item
semi-quantitative Food Frequency Questionnaire (FFQ).
A total of 3267 participants attempted and returned
the FFQ, of which 2897 were usable (79.3% of the
participants examined; 88.6% of those attempted to
complete the FFQ). Non-participants were of a similar
age and gender distribution, but had a lower self-reported
prevalence of spectacle wearing and hypertension (Smith
1997).
research_2008.indd
The second BMES study followed up participants
from the original cohort, five years later, from 1997–
1999. From the original 3654 people who participated at
baseline, 2334 attended eye examinations for the second
data collection period, representing 75% of survivors (n
= 3111); 543 (14.9%) people had died, 383 (10.5%) had
moved and 394 (10.8%) refused to participate. Of these,
2108 attempted the food frequency questionnaire, of
which 2005 were usable (85.9% of those examined).
The third BMES study followed up participants
from the baseline cohort, ten years after the original
data collection, in 2002–2004. From the 3654 people
who participated at baseline, 1952 attended interviews
for the third data collection period, representing 75.6%
of survivors. Of those who attended interviews, 1600
attempted the food frequency questionnaire, of which
1532 were usable (78.5% of those examined).
The Food Frequency Questionnaire
Dietary data were collected using a 145 item selfadministered FFQ, modified for Australian diet and
vernacular from an early FFQ from Willett & others
(1988), and included portion size estimates, as well as
frequency, strength, brand and type of supplements.
Participants used a 9-category frequency scale to indicate
the usual frequency of consuming individual food items
during the past year. Participants were asked to include
brand and type of certain foods, in particular breakfast
cereals, margarine and oil. There were no specific
questions about ripeness of fruit consumed, which can
influence the nutrient content of most fruits. However,
an allowance for seasonal variation of fruit and vegetables
was made during analyses by weighting seasonal fruits
and vegetables.
The FFQ was validated against three separate 4-day
weighed food records, collected over the period of one year
to account for seasonal variation, in a randomly selected
sub-sample of the BMES cohort (n = 79). Validation
of the FFQ has been described in peer-reviewed papers
(Smith & others 1998, Flood & others 2004).
Data entry of the FFQ
The data from each FFQ was entered into a purpose built
form in Microsoft Access. This form creates a spreadsheet
of data with subjects ID in rows and FFQ items (coded
with an item number) in the columns. A nine category
frequency scale for each food is coded with numbers 1
(never) to 9 (4+/day) for each individual. Specific food
types, such as margarine and breakfast cereals are given a
precise code which is used in the corresponding nutrient
database.
Data cleaning of FFQ
Participants with more than 25 missing values (17%)
in their FFQ were excluded from the final data set.
Participants with 13–25 missing values in the FFQ were
checked for data entry errors, which were corrected
where possible. If, after data checking, more than
12 FFQ questions remained blank or an entire page
remained blank, these FFQs were also excluded from the
final data set. Participants with FFQ daily energy intakes
of less than 2500 kJ or greater than 18 000 kJ were
excluded from the final data set. Finally, nutrient data
were screened for extreme values, inspecting values in
the upper and lower 2% of the distribution to locate and
correct any data entry errors and to check for plausibility.
This process has been repeated for each data collection
period.
Dietary analysis software
Dietary intakes in this study were estimated using a
purpose-built software package, “the FFQ Masher”
(Lazarus 1999), based on a Dbase software system. In
the initial testing of this system, the nutrient values for
a sub-sample of completed FFQ generated using the
FFQ Masher were compared to values generated from a
commercially available nutrient analysis package, Diet/1
(Xyris 1988), using an FFQ template created in that
package. Any errors identified were corrected and the
purpose built system has been used for the remaining
analyses.
Compiling the nutrient database
The compilation of nutrient databases for epidemiological
research poses many challenges, in particular, ensuring
the use of reliable and valid data (Baghurst & Baghurst
1990).
The FFQ Masher incorporated the electronic form
of the nutrient database for use in Australia (Nuttab90
in BMES1 and Nuttab95 in BMES2 and BMES3)
(Department of Community Services and Health, 1990,
1995) to which additional nutrient values for the 145 food
items were added. The Nuttab values represent analytical
data of Australian food products. However, there were
several nutrients not included in that database for which
we sought further information and so we explored
additional nutrient databases. Our preference was to
use Australian analytical data, and where these were not
available to use analytical data from other sources, with
a focus to use national nutrient databases (ie borrowed
data). However, some data used were based on a mixture
of analysed, calculated, imputed or borrowed data.
For example, folate values for foods were added from
the 1995 Australian National Nutrition Survey database,
AUSNUT (ANZFA, 1999); data on folate values of
fortified products available in Australia from 1997–1999
and then 2002–2004 were also included (Abraham &
Webb, 2001, Lawrence & others 1999; FSANZ 1999,
2004). Since fortification of foods was progressively
changing during the period of data collection, a series of
seven different time periods was identified to categorise
the folate values of fortified foods on the market. This
information was a combination of data obtained from
FSANZ and manufacturers. Each individual’s intake
was then matched to the corresponding time period in
which his or her FFQ was collected. Folate values for
foods for BMES1 reflect pre-fortified values; most of the
pre-fortified folate values used in Australia at that time
were derived from UK (Holland & others 1991) data
and did not reflect analytical data of Australian foods.
The McCance and Widdowson tables from the UK
use a single enzyme microbiological assay to determine
folate, but the preferred method now receiving support
in Australia is the tri-enzyme analytical technique which
generally yields higher folate values than the single
enzyme method (Arcot & Shreskha 2006). Additionally,
it is probable that there are real differences between
products analysed in the UK to those in Australia
research_2008.indd
(Leemhuis & others 2006).
Separating the data by time period allowed the
analyses of pre-fortified foods, so calculation of dietary
folate equivalents was possible. An allowance for the
increased bioavailability of folate from supplements
and fortified food products was calculated, expressed as
Dietary Folate Equivalents (DFE) in accordance with
US recommendations as follows: 1 µg DFE = {µg food
folate + (1.7 x µg synthetic folic acid)}(IOM 1998). The
difference between the pre-fortified folate estimate and
the fortified folate estimate was taken, and supplement
data was added to this, and then multiplied by 1.7. The
pre-fortified value was then added back to this new
figure, providing an estimate of DFE from diet and
supplements.
Vitamin B12 nutrient estimates were obtained from
the UK Composition of Foods (Holland & others 1991),
and were then added to the nutrient database used in the
FFQ Masher. There are likely to be differences between
the composition of UK foods and the true content
of Australian foods; the differences could result from
variation of food varieties, climate, growing conditions,
food manufacturing techniques and preparation
(Scheelings 1996). The UK Composition of Foods
includes foods stored in varying conditions (Holland
& others 1991). Acid, alkali, light and oxidising agents
can destroy vitamin B12, and approximately 30% of its
activity is lost during cooking (Krause & Mahan 1984).
However, in the absence of any other database, the UK
tables provided an estimate of food composition in
Australia.
The major health outcome of interest in this study
has been eye disease and subsequently there has been
a particular interest to estimate the intakes of various
carotenoids (Manzi & others 2002, Flood & others 2002,
Tan & others 2007). Where we had Australian carotenoid
data in our national electronic database, we preferentially
used this, because it was considered more comprehensive
and accurate for Australian foods. However, after this
we used the United States Department of Agriculture
Carotenoid (USDA 1998) database to estimate other
carotenoids, including -carotene, -cryptoxanthin,
combined lutein and zeaxanthin, and lycopene. It should
be noted that the 1998 USDA Carotenoid database
included the combined lutein and zeaxanthin estimates.
Later databases have separated these carotenoids using
High Performance Liquid Chromatography (HPLC)
laboratory analysis.
Food items were matched by food descriptor and,
where available, information about macronutrient intake
of a food. However, not all the items on our FFQ could
be exactly matched with one of the items in the USDA
carotenoid database. For example, our questionnaire
uses cooked (boiled) or baked (with added oil) pumpkin,
and boiled silverbeet, and the closest matches on the
USDA database were canned pumpkin, and boiled
spinach, described as ‘Spinach, cooked, boiled, drained
without salt’. The database does not list English spinach,
simply spinach. In these examples the -carotene content
of the food items from the two databases were very
different (for example spinach (US carotenoid database
5242 µg/ 100 g), silverbeet, boiled (Australian Nuttab
database 410 µg/ 100 g); ‘pumpkin, canned without
salt’ (USDA Carotenoid database 6940 µg/ 100 g),
‘pumpkin, unspecified type, peeled, boiled (Australian
Nuttab database 2676 µg/100 g)). However, as previously
indicated, we used the Australian -carotene content of
the food item. Many other items were closer in values
to the Australian database; these are the most extreme
examples where differences occurred. Differences such as
these in food preparation, food varieties and possible food
fortifications (from colouring, for example) are likely to
limit the accuracy of our estimates for the Australian food
supply and our population (Manzi & others 2002). A
useful addition to the database would have been to search
for other published analytical carotenoid data, but this
was not done at that time, as it was considered valuable
to use the relatively comprehensive USDA database. In
addition, the food preparation methods of foods may
influence the biological availability of carotenoids, in
particular for fruit and vegetables. The FFQ described
fruit as ‘fresh’ and in these cases fruit intake were assumed
to be of raw fruit. Some particular fruits in the FFQ
description, specified dried or canned fruits, which were
then assessed appropriately from the nutrient database.
The majority of vegetables were assumed to be cooked
and generally the nutrient database option was ‘boiled’.
However, there were a few exceptions: for vegetables
commonly consumed as raw items, such as lettuce,
celery, mushrooms and shallots ‘raw’ nutrient data was
used. An exception was carrot which was analysed as
50% boiled and 50% raw of unspecified carrot type.
The original fatty acid profile of nutrients in our
database was limited to saturated fats, monounsaturated
fats, and polyunsaturated fats as provided in the
NUTTAB databases. Later, we added data from the
NUTTAB fatty acid supplement database, but found
that database limiting as it had too few decimal places
for fatty acids found in very small amounts in foods, such
as the long-chain omega-3 fatty acids. Subsequently,
we used an alternative source of fatty acid data from
the Royal Melbourne Institute of Technology (Mann
& others 2002), which provided data in gram units to
two decimal places, and included an estimate for trans
fatty acids. While data described to two decimal places
may have some limitations from the point of view of
the biological variability of food sources, it has been
useful when determining the estimated nutrient intake
of some fatty acids which are available in only small but
important amounts in foods, such as long-chain omega3 fatty acids. This new data was added to appropriate
foods in the database by matching other macronutrient
content, in order to provide a close match for the food
item, since it could not be matched by ID code. However,
it is acknowledged that there are likely to be limitations to
the accuracy of the data using this system.
Dietary supplement intake
Dietary supplement intake was assessed by questionnaire,
which sought details about whether a supplement was used,
and if so, the frequency, brand and strength. Supplement
data were coded using an updated supplement database,
previously developed for use in BMES 1 (Ashton &
others 1997), which was then updated for each data
collection period. Each nutrient’s supplement data for
each participant reflects the addition of all supplements
used by that participant.
Interpretation of the data
research_2008.indd
Previous literature has documented the difficulties with
using nutrient data from FFQ to provide estimates
of absolute intake, but generally they can be used to
rank order quintiles of intakes, and our own validity
analyses have also confirmed this to be the case (Smith
& others 1998, Flood & others 2004). In addition, using
different nutrient database sources from other countries
and different methods of laboratory analysis provides
additional potential for measurement error. However, if
data are analysed by ranking individuals into groups and
then investigating these in relation to a health outcome,
then the effect of this potential measurement error is
likely to be reduced. An additional technique is not only
to consider nutrients and their relationship to a health
outcome but also to investigate related foods to a health
outcome of interest. We have been able to successfully do
this on a number of occasions and it has been a useful
way of investigating associations of diet with disease.
For example, in a paper investigating fatty acid intake
and macular degeneration, we investigated not only fatty
acid types but also major foods contributing to those
fatty acid types (Chua & others 2006). In this case, both
forms of analysis supported each other; we found those
who had the highest quintile of omega-3 fatty acid intake
versus the lowest quintile had a 60% reduced risk of
developing early macular degeneration after five years
(OR 0.41, 95% CI 0.22–0.75). In the same research we
found those who consumed fish once a week had a 40%
reduced risk of developing early AMD (OR 0.58, 95%
CI 0.37–0.9).
Lessons learned
If given an opportunity to re-create our nutrient database
for this cohort, there are several aspects of the database
we would consider doing differently:
1. Create the database using a software platform which
was more likely to allow easier modifications into
the future. The initial data entry was conducted
using Access forms. In hindsight, it would have been
easier to create the entire nutrient analyses using that
database, rather than switching to Dbase and using
the Borland Database Engine, which has become
difficult to support over time. While we are still able
to use this software it is not easy to make additions to
the system.
2. Explore other published sources of nutrient data,
particularly from laboratory analysed data of
Australian food sources. We tended towards national
nutrient databases which is likely to have limited our
use of some valuable data.
3. Document the details and exact source of all added
nutrients for each individual food item. As we
progressed with additions we learnt to specifically
document all sources of information and any known
details. Some earlier versions of additions were only
described in general terms.
Conclusions
Over time we have created a comprehensive food and
nutrient dataset to examine health outcomes over the
long period of data collection. The use of a purposebuilt software program to analyse our dietary data
has provided some flexibility with including additional
nutrients to the dataset. We have taken care about the
approach to consider additions to reflect changes to the
food supply, such as fortification, and to consider how
we can use nutrients from other nutrient datasets. It is
important to remember the limitations of the dataset,
generally using the data in a ranked form, interpreting
both nutrient and food consumption. Anticipating the
changing needs of researchers involved with the study
also adds to the utility of a nutrient database, and ideally
this should be considered in the planning stages of
database construction.
References
Abraham, B & Webb, K. 2001. Interim evaluation of the voluntary folate
fortification policy. Australian Food and Nutrition Monitoring
Unit. AusInfo, Canberra.
Arcot, J & Shrestha, A. 2005. Folate: methods of analysis. Trends in
Food Science and Technol. 16: 253-266.
Ashton, BA & others. 1997. Development of a dietary supplement
database. Aust. NZ J Pub. Health 21: 696-702.
ANZFA. 1999. National Nutrition Survey Nutrient Database. Australia
and New Zealand Food Authority. Canberra.
Baghurst, KI & Baghurst, PA. 1990. The uses of food compositional
data in nutritional epidemiology. In (Greenfield, H. ed). Uses and
abuses of food composition data. Food Aust. 42(8): 511-513.
Chua, B & others. 2006. Dietary fatty acids in 5-year incidence of agerelated maculopathy. Arch Ophthal. 124: 981-986.
Chug-Ahuja, JK & others. 1993. The development and application of
a carotenoid database for fruits, vegetables and, selected multicomponent foods. J Am. Diet. Assoc. 93 (3): 318-23.
Department of Community Services and Health. 1990. Nuttab90.
Nutrient Database for use in Australia. Australian Government
Publishing Service, Canberra.
Department of Community Service and Health. 1995. Nuttab95.
Nutrient Database for use in Australia. Australian Government
Publishing Service, Canberra.
Flood, V & others. 2002. Dietary antioxidant intake and incidence
of age-related maculopathy: the Blue Mountains Eye Study.
Ophthalmology. 109: 2272-8.
Flood, VM & others. 2004. Issues in assessing the validity of nutrient
data obtained from a food frequency questionnaire: folate and
vitamin B12 examples. Pub Health Nutr. 7: 751-756.
Holland B & others. 1991. McCance and Widdowson’s The Composition
of Foods. 5th Ed. The Royal Society of Chemistry, Cambridge and
Ministry of Agriculture, Fisheries and Food, London.
IOM. 1998. Dietary references for thiamin, riboflavin, niacin, vitamin
B6, folate, vitamin B12, pantothenic acid, biotin and choline.
Institute of Medicine, National Research Council. National.
Academy Press, Washington, DC.
Krause, MV & Mahan, LK. 1984. Food, Nutrition and Diet Therapy.
7th ed. W.B. Saunders Co, Philadelphia USA.
Lawrence, MA, Rutishauser, IHE & Lewis, JL. 1999. An analysis of
the introduction of folate-fortified products into stores in Australia.
Aust. J Nutr. & Diet. 56: 15-21.
Lazarus, R. 1999. FFQ Masher: Food frequency questionnaire software
analysis package.
Leemhuis, C, Cunningham, J & Arcot, J. 2006. Comparing Australian
and United Kingdom folate values. Food Standards Australia New
Zealand. Poster.
Mann, NJ & others. 2003. Development of a database of fatty acids in
Australian foods. Aust. J Nutr. & Diet. 60: 42-45.
Manzi, F & others. 2002. The intake of carotenoids in an older
Australian population: The Blue Mountains Eye Study. Public
Health Nutr. 5: 347-352.
Scheelings, P. 1996. Prospects for measuring folate in Australian foods.
Aust J Nutr. & Diet. 53S: S23-S28.
Smith WT. 1997. PhD thesis: Risk factors for age-related maculopathy.
1-287.
Smith, W & others. 1998. Validity and reproducibility of a selfadministered food frequency questionnaire in older people. Aust.
NZ J. Publ. Health 22: 456-463.
Tan, J & others. 2007. Dietary antioxidants and the long-term incidence
of age-related macular degeneration: Blue Mountains Eye Study.
Ophthalmol. In Press. On-line doi: 10.1016/j.opththa.2007.03.083
USDA. 1998. USDA – NCC Carotenoid database.
United States Department of Agriculture http://www.nal.usda.gov/fnic/
foodcomp/Data/Carot/.
research_2008.indd
Willet, W & others. 1988. The use of a self-administered questionnaire
to assess diet four years in the past. Am. J. Epidemiol. 127: 188199.
Xyris. 1988. Software. Diet 1.3. Dietary analysis software. Brisbane.
Dr Victoria Flood is Nutritional Epidemiologist with the NSW Centre for
Public Health, Nutrition, Human Nutrition Unit, University of Sydney,
2006 and the Centre for Vision Research, Westmead Millennium Institute,
University of Sydney, 2145; email: [email protected]. Professor
Wayne Smith is with the Centre for Clinical Epidemiology and Biostatistics,
University of Newcastle, NSW 2300. Elena Rochtchina, Senior Statistician,
Dr Jie Jin Wang is Deputy Director and Professor Paul Mitchell is Director
with the Centre for Vision Research.